What does this skill do, and when should you use it?
Launch Retro Analyzer runs a structured D1, W1, or M1 retrospective after a product launch. It treats UTM-attributed analytics from the team’s own data as the truth column and keeps platform-reported figures in a separate reference column, with every number labeled by provenance. The skill runs one 5-Whys chain on the largest miss, makes a keep/kill/change call for every channel, and drafts 3–5 actionable learnings for the next launch. It can submit an outcome snapshot through an authorized proposal, but it does not write launch-registry records directly.
Reads preregistered KPI targets, accepted launch context, T-0 through T+30 tracking, own UTM-attributed analytics, and separately labeled platform dashboards; builds a per-channel actual-versus-target table with Measured, User-provided, or Estimated labels; selects the single largest miss and runs a change-oriented 5-Whys chain of up to five levels; produces a keep, kill, or change decision with a one-line reason for each channel; drafts 3–5 actionable next-launch learnings; submits a launches.ndjson outcome-snapshot proposal to registry-events.py when authorized; and emits the standard handoff summary.
- A growth lead has completed a product launch and needs a W1 comparison of channel performance against declared KPIs.
- A team has one major channel or KPI miss and needs a focused root-cause chain instead of a separate deep dive for every row.
- A launch review requires evidence-based keep, kill, or change recommendations using a GA4 or equivalent UTM export.
- A team is closing a launch and needs checkable learnings plus an outcome snapshot for the launch registry.
How do you install this skill?
- The skill was not executed; static documentation does not verify attribution, 5-Whys quality, or registry-submission success.
- Before use, confirm that GA4/UTM exports are available, targets were actually preregistered, and all proposed event contents are reviewed.
- The declared skill version 18.0.0 is inconsistent with the security policy's stated supported 17.x line; verify maintenance and security coverage.
- Although connectors are described as optional, separately verify applicable terms, data egress, and mainland-China network reachability.
- Shell / CLI
- Network access
- Local filesystem
Python standard-library connector scripts
Install the repository with an Agent Skills-compatible host: npx skills add aaron-he-zhu/aaron-marketing-skills. You can also run git clone https://github.com/aaron-he-zhu/aaron-marketing-skills. The skill is at launch/prove/launch-retro-analyzer/SKILL.md. For Claude Code, the source provides: /plugin marketplace add aaron-he-zhu/aaron-marketing-skills, followed by /plugin install aaron-marketing@aaron.
tmp="$(mktemp -d)"
git clone --depth 1 https://github.com/aaron-he-zhu/aaron-marketing-skills.git "$tmp"
mkdir -p ~/.claude/skills
cp -R "$tmp/launch/prove/launch-retro-analyzer" ~/.claude/skills/
rm -rf "$tmp"Generated from the source repository and skill path; it copies only this skill's folder. If the author's install steps above differ, follow those first. To scope it to one project, replace ~/.claude/skills with that project's .claude/skills.
How do you use this skill?
Provide the product launch, retrospective window, declared targets, and analytics export. Example: Run a W1 retro on our [product] launch. Targets: [D0/W1 KPIs]. Here is the GA4 UTM export and the platform dashboards. Another supported prompt is: Close out the [product] launch: build the actual-vs-target table, log the learnings, and submit the outcome snapshot to the launch registry. If targets are missing, the retro should be marked NEEDS_INPUT; saving results requires confirmation first.
What are this skill's strengths and limitations?
- Keeps first-party UTM-attributed actuals separate from platform self-reported reference figures.
- Provides a constrained, repeatable output: channel table, one 5-Whys chain, channel decisions, and next-launch learnings.
- Works at Tier 1 with pasted or exported data, without requiring paid tools or MCP.
- Treats exports and pasted reports as untrusted input and restricts registry-bound writes to authorized proposals.
- A complete retro depends on declared targets; missing targets may force a NEEDS_INPUT outcome.
- It does not calculate CPA, ROI, or payback, write stakeholder-facing reports, perform metric deep dives, or monitor the live launch window.
- Platform-mechanic explanations remain Estimated hypotheses with a named source rather than confirmed root causes.
- The source does not provide independent test results or platform-by-platform validation for this individual skill.
How does this skill compare with similar options?
Side by side with related skills; every score comes from the same FSRS standard.
| Skill | FS score | Stars | Last updated | License |
|---|---|---|---|---|
| Launch Retro Analyzer this page | 59 · Recommended | ★ 3k | 12d ago | Apache-2.0 |
| RAMP Launch Readiness Auditor | 61 · Recommended | ★ 3k | 12d ago | Apache-2.0 |
| Launch Pricing & Packaging Planner | 61 · Recommended | ★ 3k | 12d ago | Apache-2.0 |
| Community Launch Runner | 56 · Use with care | ★ 3k | 12d ago | Apache-2.0 |
| Launch Tier Planner | 63 · Recommended | ★ 3k | 12d ago | Apache-2.0 |
launch-monitor covers upstream T-0 through T+30 tracking, while this skill performs the downstream D1/W1/M1 retrospective. Use roi-calculator for CPA, ROI, and payback math; report-generator for stakeholder-facing reporting; and performance-analyzer for metric deep dives or anomaly analysis.
How did FollowSkills review this skill?
The skill treats exports as untrusted input, asks for confirmation before persistence, and limits registry interaction to an authorized operation: propose request. It documents data flow, write boundaries, keyless defaults, and external effects. Points are deducted because event submission and local memory writes still create side effects, while the skill itself lacks a complete rollback procedure, permission configuration, and detailed sensitive-data minimization rules.
Instructions, inputs, outputs, and scope boundaries are largely self-consistent, with NEEDS_INPUT for missing targets and explicit rejection of embedded instructions. Points are deducted because the supplied static material does not reproduce the skill's key paths, runtime dependency availability is not demonstrated, and abnormal-input failure feedback is only lightly specified; static calibration caps this at 10.
Triggers, D1/W1/M1 windows, inputs, outputs, exclusions, and handoff targets are clearly described, with Chinese labeling and a core path based on user-provided data rather than mandatory overseas services. Points are deducted because non-fit cases such as absent UTM data or conflicting multi-channel evidence are not fully specified, and mainland-China environment support is not independently evidenced.
Frontmatter, Quick Start, Skill Contract, Data Sources, Instructions, Save Results, and Next Best Skill provide useful progressive disclosure; Apache-2.0, a version, and security-policy material are present. Points are deducted because maintenance ownership, changelog details, and update responsibility for this skill are unclear, and the security policy references supported 17.x while the skill declares 18.0.0.
The expected deliverables are concrete and directly usable: per-channel actual-versus-target tables, one 5-Whys chain, keep/kill/change decisions, learning entries, and an outcome snapshot. Points are deducted because no representative output or third-party execution evidence is provided, so attribution quality, successful registry submission, and marginal benefit over manual work cannot be verified statically; static calibration caps this at 7.
The skill cites shared protocol, security, and connector documentation, while repository architecture, validator tests, and CI provide some auditable support. Points are deducted because the tests do not cover this skill's key path and there is no independently reproducible execution result or corroborating third-party evidence; static calibration caps this at 5.
Open a dimension to read why it scored that way
Evidence confidence：Low — Mostly static review, author material or a limited demo; useful for discovery, not high-risk decisions.
See the full review method →